© 2004 Plant Management Network.
Developing Phenological Prediction Tables for Soybean
Lingxiao Zhang, Delta Research and Extension Center, Jiuquang Zhang, Computer Science Department, Clarence E. Watson, Mississippi Agricultural and Forestry Experiment Station, and Stephen Kyei-Boahen, Delta Research and Extension Center, Mississippi State University, Stoneville 38776
Corresponding author: Lingxiao Zhang. firstname.lastname@example.org
Zhang, L., Zhang, J., Watson, C. E., and Kyei-Boahen, S. 2004. Developing phenological prediction tables for soybean. Online. Crop Management doi:10.1094/CM-2004-1025-01-RS.
Accurate prediction of soybean growth and development stages is important for soybean farmers and researchers to effectively manage production practices. The objective of this study was to develop accurate time-tables to predict soybean growth and development stages under field conditions. Data from experiments conducted under irrigated conditions at Stoneville, MS from 1998 to 2002 were used. The period of prediction covered the entire growing season from early March to late October, and maturity group (MG) ranged from early III to late V. Tables were constructed based on regression analysis using iterative approximation. Cross validation was performed using randomly selected subsets of the 2002 data set. Accuracies of prediction were similar to model fitting errors. Overall, the error of prediction ranged from 0 to 8 days. It was concluded that soybean phenological stages could be adequately predicted using regression approaches and tables and methodology of table construction can be used by other researchers to build models and tables for their particular situations.
Soybean (Glycine max (L.) Merrill) growers often need to schedule production practices (e.g., irrigation, weed control, and harvest) based on their knowledge of soybean growth and development. Producers, especially multi-crop producers, often need to schedule labor and machinery use ahead of time to insure efficient utilization. Timing of soybean growth and development events, such as flowering and maturity, can directly affect grain yield (9,11). Therefore, an understanding of soybean phenological events is essential for research and extension personnel to provide proper guidelines and assistance for producers. Since soybean is a photoperiod-sensitive plant, its growth and response to photoperiod may differ at two locations if latitude, longitude, and/or elevation differ significantly. For soybean researchers to accurately describe soybean phenological development, it is important to use standard terminology to describe growth events. Soybean developmental stages were systematically defined by Fehr and Caviness (2). Their system has been adopted as a standard and is widely used in soybean phenological description.
Soybean phenological developments are influenced by numerous factors, and thus the length of the entire growth period and the length of individual growth stages from year to year are not constant for a particular cultivar. Factors influencing phenological development include MG, planting date, and environmental stresses (water, temperature, nutrition, etc.). Many elaborate models have been developed to predict soybean phenological development (3,4,5,6,8,11); however, growers can hardly use most of these models due to the complexity and specificity of the models (1).
The history of soybean production in Mississippi is relatively short compared to most Midwestern states. Previously, most soybeans grown were late MGs (such as MG VI, VII, or VIII) planted from late May to late June. In recent years, early-maturing cultivars are popular in the Mississippi Delta and the Mid-South. A recent survey indicated that 93% of the soybean acreage in Mississippi was planted to MG IV and V cultivars (13). Planting dates have moved earlier during this same period. As a result, there is little or no systematic information available about soybean phenological development under these conditions.
Historical field data can be utilized to estimate the approximate dates of specific growth events through simulation with a reliable crop growth modeling procedure; however, a simplified method, such as a growth stage table, may be more realistic and easier to use for producers and county agents. The tables may help to explain the complex relationships between soybean growth and response to environmental conditions.
On-line references are widely used in agricultural research and extension; however, many farmers and extension agents prefer printed documents for field use. The objective of this study was to develop simple time-tables for growers and extension agents to use in growth stage prediction based on regression models developed from field data.
Experiments were conducted at the Delta Research and Extension Center at Stoneville, Mississippi (33.4°N, 90.9°W) from 1998 to 2002. Planting dates ranged from early March to early July and were broadly categorized into eight groups by the following time frames: early March, late March, early April, late April, early May, late May, early June, and late June/early July (Table 1). During the study period, different planting dates were used each year due to environmental conditions at planting time. The experimental design was a split-plot with planting dates as main plots and cultivar as subplots arranged in a randomized complete block design with three or four replications. Soybean cultivar from MG III to MG V were evaluated in all years (Table 2).
Table 1. Summary of planting dates, 1998-2002.
† n.a. = Not applicable, there was no planting.
Table 2. Soybean cultivars and maturity groups evaluated during 1998 through 2002.
Plots consisted of four rows 20 inches wide and 30 ft in length. The field was irrigated using an overhead sprinkler system starting in late May or early June to minimize water stress. The timing of irrigation was determined whenever initial leaf wilting was visually observed. Weeds were controlled by cultivation between the 3- to 4- leaf stage and with herbicide (Classic, etc.) application as necessary.
The date when each cultivar reached a phenological stage (2) was recorded by observing plants in each plot every other day. Data for each maturity group were analyzed separately for each year using general SAS (SAS Institute, Cary, NC) linear model procedures. The error terms for each year were examined for homogeneity of variance using Bartlett’s test, and then combined analyses across years were performed for each MG.
Several regression analyses were performed using SAS procedures and the linear model was selected as the best fit. One model was built for the vegetative growth stages with planting date (day of year) and normalized growth stage as predictor variables, whereas separate models were constructed for the reproductive stages using MG and planting date as predictor variables. The data set from the 8 May 2002 planting date was randomly selected to validate the models. Cross validation was performed using subsets with 34 observations (n = 34) each for the vegetative growth model and 27 observations (n = 27) each for the reproductive stage models. An iterative approach was used in the validation procedure. With this approach, the optimum coefficients for planting date, growth stage, or MG were estimated n different times using n-1 observations. On each iteration, a different observation was left out, and the predicted date and prediction error were computed for this particular observation. This step was repeated for each observation and a set of n coefficients and prediction errors were estimated. The average of these values were then computed.
An interpolative approach was then employed to estimate the dates for the growth stages between the known data sets. This produced a table containing predicted dates of growth stages for all combinations of planting date and maturity group within the range of data tested. Since this table is intended for use by growers and extension agents to predict soybean growth stages, the data in the tables are presented in 5-day intervals with maturity group intervals of 0.5 from maturity group values ranging from 3.4 to 5.9.
Time-Tables for Soybean Growth and Development Stage
Data analysis indicated no significant differences among MGs for early vegetative growth (up to 8th node) within the same planting date; therefore, data was pooled across MGs (data not shown) (Appendix, Table 3). Significant MG interactions were detected during the reproductive growth stages, therefore results were separated. Dates for reproductive growth events of different MGs for different planting dates are summarized in Table 4 (Appendix).
A cross validation procedure (7,10) was carried out comparing results of the phenological tables with observed field data selected randomly from the data set for the 8 May 2002 planting date. Test results indicated that the correlation between estimated dates and observed dates in the field were very high (R2 > 0.95). Figure 1 illustrates the results of the cross validation in terms of the goodness of fit. For the vegetative stage, the predicted dates for the early stages (VE to V4) were late compared with field observations. In contrast, dates predicted for V6, V7, and V8 were earlier than observed with a maximum prediction error of 6 days, which occurred at V8. The model predicted early dates for the reproductive stages, and the error increased after R5. The maximum error for the reproductive stage was 8 days at R7. The temperature for May 2002 was on average 2.4°F lower than the same period during the four preceding years and this possibly delayed the early vegetative growth. However, the differences between the predicted and observed dates after V5 through R8 could not be accounted for since temperature and moisture conditions for the rest of the growing season were similar to those for the preceding years.
Some soybean growth stages are hard to distinguish under field conditions on a single-day basis. The primary use of these tables will most likely be to predict maturity date of soybean cultivars planted on a specific date. Cross validation analysis indicated that the difference between observed and predicted dates when using this model to predict the growth stages could range from 0 to 8 days; however, the frequency of getting 8 days difference at maturity (R8) was very limited. Therefore, it should not affect the usefulness of these tables too much.
In this study, only MG and planting dates were considered input variables to predict phenological events. Since these data were obtained from irrigated plots, it was assumed that the effect of water stress was minimized. On non-irrigated fields, the periods required for each developmental stage are generally shorter with the degree of reduction dependent upon environmental factors such as water deficit and temperature stress. This suggests that these tables are best suited for irrigated fields and caution should be used if one attempts to apply them to fields that may have suffered from environmental stresses such as water and temperature.
Since these tables were derived under field conditions at Stoneville, MS, it should be suitable for predicting phenological development of soybeans grown at locations with similar latitude. In this study, temperature was not a predictor factor. Thus, greater variation in prediction is likely with extremely early plantings (prior to March 25) due to the variable temperature effects in early spring, which may greatly affect the initial growth rate of the plant. Variation may also be greater with late maturing cultivars (MG VI or later) since cool temperature and excessive precipitation during the later part of the growing season may delay maturity.
In conclusion, the 5-year study provided an excellent data set for predicting soybean growth stages using regression models. The approach predicted the growth stages with reasonable accuracy; however, because the data sets for the prediction were obtained under irrigated conditions at Stoneville, MS, it may not be reliable for soybean grown on non-irrigated fields or for locations outside this latitudinal zone. The accuracy of prediction for soybean planted too early (generally before 25 March) or for late maturing cultivars may be affected because cool temperatures and excessive rainfall that occur around that period often reduce the initial growth rate or delay maturity.
1. Boote, K. 1997. Gossym/Comax: The quixotic quest. Page 15 in: Biological Systems Simulation Group (Abstr.) Agric. Biol. Engr. Dep., Univ. of Florida, Gainesville.
2. Fehr, W. R., and Caviness, C. E. 1977. Stages of soybean development. Iowa Agric. Exp. Stn. Special Rep. 80.
3. Grimm, S. S., Jones, J. W., Boote, K. J., and Hesketh, J. D. 1993. Parameter estimation for predicting flowering dates of soybean cultivars. Crop Sci. 33:137-144.
4. Grimm, S. S., Jones, J. W., Boote, K. J., and Herzog, D. C. 1994. Modeling the occurrence of reproductive stages after flowering for four soybean cultivars. Agron. J. 86:31-38.
5. Hesketh, J. D., Elmore, C. D., and Jones, J. W. 1980. Predicting flowering and subsequent leaf expansion. Pages 123-132 in: Predicting Photosynthesis for Ecosystem Models. J. D. Hesketh and J. W. Jones, eds. CRC Press, Inc., Boca Raton, FL.
6. Hodges, T. 1990. Temperature and water stress effects on phenology. Pages 7-13 in: Predicting Crop Phenology. T. Hodges, ed. CRC Press, Boca Raton, FL.
7. Irmak, A., Jones, J. W., Mavromatis, T., Welch, S. M., Boote, K. J., and Wilkerson, G. G. 2000. Evaluating methods for simulating soybean cultivar response using cross validation. Agron. J. 92:1140-1149.
8. Jones, J. W., Boote, K. J., Hoogenboom, G., Jagtap, S. S., and Wilkerson G. G. 1989. SOYGRO V5.42. Soybean crop simulation model: User’s guide. Fla. Exp. Stn. Journal No. 8304. Agric. Biol. Engr. Dep., Univ. of Florida, Gainesville.
9. Laing, D. R. L., Jones, P. G., and Davis, J. H. C. 1984. Common bean (Phaseolus vulgaris L.). Pages 305-351 in: The Physiology of Tropical Field Crops. P. R. Goldsworthy and N. M. Fisher, eds. Pudoc, Wageningen, the Netherlands.
10. Mitchell, T. M. 1997. Machine Learning. McGraw-Hill Company, Boston, MA.
11. Pedersen, P., Boote, K. J., Jones, J. W., and Lauer, J. G. 2004. Modifying the CROPGRO-soybean model to improve predictions for the Upper Midwest. Agron. J. 96:556-564.
12. White, J., and Izquierdo, M. 1991. Physiology of yield potential and stress tolerance. Pages 287-382 in: Common Beans: Research for Crop Improvement. A. van Schoonhoven and O. Voysest, eds. CAB Inter., Cambridge, MA.
13. Zhang, L. X., Poston, D., and Cooke, F., Jr. 2001. Survey of recent soybean production trends in the Mississippi Delta. 2001 Annual Meeting Abstracts. CD-ROM. ASA, CSSA, and SSSA, Madison, WI.
Table 3. Date of attainment of vegetative growth stages (VE to V8) by planting date (PD) averaged across 1998-2002 and across soybean maturity groups under field conditions at Stoneville, Mississippi.
Table 4. Date of attainment of reproductive growth stages (R1 to R8), by maturity group (MG) and planting date (PD) for soybeans under field conditions averaged across 1998-2002 at Stoneville, Mississippi.